PIDD-MSK: Physics-Informed Data-Driven Musculoskeletal Modelling

点击翻译按钮获取中文摘要

Musculoskeletal models are powerful, computational simulation tools for detailed biomechanical analysis to estimate internal movement variables, which cannot be readily measured in vivo but are of interest to clinicians. The traditional physics-based musculoskeletal models are time-consuming and with high running latency, while the data-driven musculoskeletal models are fast but cannot reflect the underlying physical mechanisms.This proposal is to overcome limitations associated with both types of musculoskeletal models while preserving their advantages, and my vision is to develop the next-generation physics-informed data-driven musculoskeletal models, which can seamlessly integrate the existing physics-based domain knowledge into the data-driven models. The main innovations are 1) to bring physics information into data-driven musculoskeletal models to overcome its limitations by creating data reflecting the underlying physical mechanisms, and 2) to bring deep learning into physics-based musculoskeletal models to reduce computational demands in data processing and improve execution speed for real-time applications. I will apply the developed model for quantitative upper extremity impairment assessment for stroke patients as the case study. In addition to stroke patients, the outcome of this project can also potentially benefit millions of people with conditions such as multiple sclerosis and spinal cord injury, as well as people with musculoskeletal conditions.This fellowship would also improve my research skills, strengthen my academic profile, build my leadership capability and broaden my collaboration links in machine learning/deep learning for healthcare, and it will lay the foundation for me to become a worldleading expert in this area so that I can make scientific contributions and generate societal and economic impacts for Europe in the future.